ACTS

An Active Learning Method for Time Series Classification

A very basic implementation of the ACTS algorithm for Time Series Classification. Made for testing purposes. The paper is available at the following link

Made by Davide Badalotti and William Lindskog for MSc thesis at Viking Analytics.

Overview

The algorithm is constructed similarly to a modAL query strategy, except for some additional arguments.

It was originally built for another library, but will work with modAL as well.

More info on implementation at the documentation directory.

Usage

To use the algorithm first clone the repository and import the object.

from ACTS import ACTS

Then, create an instance of the ACTS class as:

acts = ACTS()

The query strategy itself is in the ___call___ function, so:

query_idxs = acts(n_instances, X, DL, L, Li)

Args

n_instances : int number of instances to be queried
X : np.ndarray of shape (n_unlabelled_data, n_points). Contains the unlabelled instances.
DL : np.ndarray of shape (n_labelled_data, n_points). Contains all the labelled instances.
L : np.ndarray of shape (n_labelled_data, ). Contains the labels of DL
Li : np.ndarray of dtype=int of shape (n_labelled_data, ). Contains the indices of the labelled instances in DL

Returns